Assessment of the impact of climate change on groundwater resources using regional climate model projections: comparison of surrogate modeling techniques

This work compares two different surrogate data-driven models in order to evaluate the effects of climate change on groundwater resources by means of an ensemble of 13 Regional Climate Models (RCM), provided within the Euro-Cordex project, under two different scenarios (RCP 4.5 and RCP 8.5). The impact was evaluated for three future periods: 2006-2035 (short term), 2036-2065 (medium term) and 2066-2095 (long term). Both approaches are based on historical data collected in northern Tuscany, covering the period 2005-2020. Historical precipitation and temperature records, observed in 18 gauging stations, and piezometric levels for 14 wells were used to build the surrogate models. The first methodology is based on a linear regression model and adopted standardized indices: the Standardized Precipitation Evapotranspiration Index (SPEI) and the Standardized Groundwater Index (SGI). First, for each well, the correlations between SPEIs and SGIs were investigated for the period 2005-2020. In case of meaningful correlation, linear regression equations are used to estimate SGIs as function of SPEIs. The linear regression models were then applied to predict future SGIs ​​using SPEIs computed from the data ​​provided by the RCMs projections. The second surrogate technique involves the use of a Long-Short Term Memory (LSTM) neural network. LSTM allows to work directly with climate variables and normalized groundwater levels. The LSTM network was trained using the historical precipitation and temperature time series for the period 2005-2018 as input and the normalized groundwater levels as output. Rain, temperature and piezometric level data from 2019 to 2020 were used to test the network. Subsequently, the rainfall and temperature time series ​​provided by the RCMs have been used by the LSTM to predict the future groundwater levels. The analysis highlights, for both approaches, a negative impact of climate change on the groundwater system. In particular, according to the RCP 4.5 in the medium-term period a larger reduction of groundwater availability is expected, while with the RCP 8.5 the long-term period is the most affected by a groundwater level decline.